INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Dyspnea (shortness of breath) is a major and disabling symptom of cardio-pulmonary disease as well as a distressing sensation frequently reported by patients suffering from anxiety and panic disorder. Dyspnea is reported as frequently as pain by seriously ill patients and thus degrades quality of life in millions of people worldwide (Desbiens et al. 1997). The general term dyspnea subsumes several sensations such as sense of respiratory work and effort, tightness of asthma, and air hunger (an uncomfortable urge to breathe). In disease, increased work, effort, or chest tightness are commonly present with air hunger, and we believe that air hunger, in particular, is important in producing the unpleasant nature of dyspnea. While important in disease states, air hunger also has a role in normal conscious behaviors that address threats to adequate breathing (for instance, long breath-hold dives or external obstruction of the airways). Similar to pain, the compelling and primal nature of air hunger has been compared with other essential vegetative sensations such as thirst and hunger (Banzett et al. 2000; Liotti et al. 2001). Air hunger is increased by afferent inputs demanding more ventilation; it is decreased by afferent inputs reporting the prevailing ventilation. For example, stimulation of respiratory chemoreceptors by CO₂ or hypoxia increases air hunger (Banzett et al. 1989, 1990, 1996) and mechanoreceptor traffic reporting tidal inflation of the lungs relieves air hunger (e.g., Hill and Flack 1908; Manning et al. 1992; Opie et al. 1959). Air hunger presumably arises from the integration of such inputs within the CNS.

Until very recently, the cerebral representation of dyspnea was entirely unknown, mainly due to an absence of useful animal models or relevant clinical lesion studies. Positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) have been widely used to study pain, showing prominent cerebral representation in limbic areas (reviewed by Casey and Bushnell 2000). A PET study of respiratory motor activation during CO₂-stimulated breathing provided an initial suggestion that limbic areas might be involved in dyspnea perception (Corfield et al. 1995). Subsequently, three H₂¹⁵O PET studies of cerebral activation in humans during laboratory-induced dyspnea have been published in five reports (Banzett et al. 2000; Brannan et al. 2001; Liotti et al. 2001; Parsons et al. 2001; Peiffer et al. 2001). Strong activation of the anterior insular cortex was common to all of these studies. Several other structures have been activated in one or two of these early studies (but not all), suggesting that loci in addition to the insula may be involved in dyspnea perception.

A principal objective of the present study was to identify neural correlates of air hunger using blood oxygen level dependent (BOLD) fMRI. This approach would enable us to conduct our primary and control experiments in the same subjects and thereby quantitatively account for aspects of the stimulus unrelated to air hunger. We used an intervention similar to that used previously in our laboratories: tidal volume (V_T) delivered by a mechanical ventilator was reduced to produce air hunger and increased to relieve air hunger while blood gas concentrations were held constant by manipulating inspired gas concentrations. We acquired brain images during a rapidly changing stimulus that optimized fMRI signal-to-noise ratio (S/N). Our control experiment employed comparable alterations in V_T but was conducted at a Pa_CO2 below the threshold for air hunger. Hypercapnia increases global brain blood flow about 5% per torr rise in Pa_CO2 (Kety and Schmidt 1948; Poulin et al. 1996; Ramsay et al. 1993b); this flow change affects the functional signal. Constant CO₂ within each study avoided changes in global brain blood flow that would correlate with experimental condition and potentially confound the interpretation. Flow increases caused by CO₂ may not be uniform across the brain (Ito et al. 2000); such nonuniformity would add noise to functional activation maps and could give rise to signal changes unrelated to neural activity.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

We studied six healthy right-handed subjects (age range 25-32 yr, 4 women). All subjects gave informed consent and were studied with ethical approval (Joint Ethical Committees of The Institute of Neurology, The National Hospital for Neurology and Neurosurgery and the Imperial College School of Medicine, Charing Cross Hospital Campus and the Harvard School of Public Health). All denied a clinical history of neurological, cardiopulmonary, and psychiatric illness, including panic disorder. Two of the six subjects were respiratory physiologists naïve to the experimental protocol. The remaining subjects had no prior experience or participation in respiratory/dyspnea research. None of the subjects was an author of the present study or participated in the prior PET study from our laboratories.

Experimental design

We induced repeated episodes of air hunger while measuring cerebral activity. During scanning, the subjects' ventilation was controlled by a clinical mechanical ventilator (Siemens 900B). The stimulus interventions and resultant sensation have been characterized in several prior studies (e.g., Banzett et al. 1989, 1996, 2000). We conducted two separate experiments in the same six subjects. Each experiment alternated between two conditions: a low V_T condition and a high V_T condition (Fig. 1); respiratory frequency was held constant (mean 13.2 breaths/min). End tidal P_CO2 (PET_CO2) was held constant by raising inspired CO₂ during high tidal volume. Eight 42-s low V_T epochs alternated with eight 42-s high V_T epochs. Changes in air hunger following a step change in volume may take up to a minute to come to a new steady state (Shea and Evans 1994); thus, if low V_T epochs had been longer, subjects would have reached severe or extreme air hunger. In the primary experiment, end tidal P_CO2 (PET_CO2) was held at 41 Torr (mean ± SD = 2.4); at this CO₂ level, the low V_T condition (V_T; mean = 0.75 ± 0.21 L) reliably evoked strong air hunger (moderate to severe) and the high V_T condition (mean = 1.38 ± 0.26 L) consistently provided relief. The second experiment served as a control for activations related to V_T change but not related to air hunger. This control experiment employed comparable levels of V_T (0.94 and 1.47 L), but was conducted at lower constant PET_CO2 (33 Torr, ± SD = 1.6), below the threshold for air hunger in either V_T condition.

View larger version (53K): [in this window] [in a new window]   Fig. 1. Psycho-physiological results from a single subject (representative of the group). Subjective air hunger ratings shown in the top panel reported breath by breath: NO, none; SL, slight; MO, moderate; SE, severe; EX, extreme. Ratings aligned with respiratory waveforms: PCO2, partial pressure of end tidal CO2; PO2, partial pressure of end tidal O2; VT, expired tidal volume; PAW, airway pressure. Repetitive provocation and relief of air hunger in response to varying tidal volume is depicted, while the end-tidal gases were held relatively constant. Uniform PAW, during periods of high VT (bottom waveform), represents consistent lung inflation, providing evidence for the subject's relaxation of the respiratory musculature. Breath-to-breath variation of PAW, during low VT, indicates increased respiratory muscle activity.

Because regular periodic changes in stimulus may lead to anticipation by subjects, additional "ruse epochs" were incorporated into both the primary and the control experiments. "Shortened" (1/3 length of standard epoch) and "prolonged" (double length of standard epoch) low V_T epochs were incorporated into the primary experiment (but these were adjusted to prevent extreme ratings). Two "very low" V_T epochs (0.6 L) were incorporated into the control experiment. These ruse epochs caused variations in the time course and intensity of sensations.

This experiment was designed to overcome challenges unique to the fMRI environment. The subject lay supine within the magnetic resonance (MR) scanner. The ventilator was located in the control suite (outside of the scanner room) and ventilation was delivered via a mouthpiece connected to 5.5 m of stiff smooth-bore tubing; inspiratory tubing was 1.6 cm ID; expiratory tubing was 2.5 cm ID. Expiratory resistance was 0.7 cmH₂O·l¹·s¹ (inspiratory resistance is borne by the ventilator). Compliance of the system was 4.6 ml/cmH₂O; thus ~10% of measured V_T was lost to gas compression (we have not applied a correction to the reported values). To avoid artifact caused by head movement, subjects breathed through an individually fitted mouthpiece, which also served as a bite-bar. For each scan series (8 epochs of alternating V_T), the amount of head movement from one whole brain scan to the next within the series was less than the width of a voxel (<0.5 mm of translation in any plane and <0.5° of rotation; assessed by SPM99 motion correction algorithm).

Physiological measurements

Airflow and airway pressure were measured by pneumotachometers and pressure transducers, integral to the ventilator (calibrated prior to each study). Arterial blood gas levels (Pa_CO2 and Pa_O2) were inferred from end expiratory gas levels sampled by a mass spectrometer (MGA 900, Case Medical). V_T was derived off-line from the airflow signal. All physiological waveforms were recorded to magnetic disk (Dell Optiplex-GX, Dell Computer) with scan event marks so that physiological variables could be correlated in time with individual brain scans.

Psychophysical measurements

Air hunger rating was reported each breath, indicated by a finger-operated keypad; this was recorded alongside physiological measurements. Because a key press was required in both test and baseline conditions, we assume that cerebral activations related to the motor act of key pressing would cancel in statistical comparisons. A standard script was used instructing subjects to rate "an uncomfortable urge to breathe." We employed a five-point scale, as follows: "Zero, Mild, Moderate, Severe, and Extreme." "Severe" was defined as an intensely unpleasant urgent urge to breathe that could be tolerated for only a short period of time and "Extreme" was defined as intolerable. Subjects were told that the stimulus would be terminated immediately if they signaled Extreme. An example of the stimulus-response relationship during an actual experiment is provided in Fig. 1.

Subjects were interviewed immediately following each experiment. They first were asked to provide a free unguided description of their sensations and later were asked to choose descriptors from a set list to detail the quality and intensity of the sensation experienced during their Moderate and Severe ratings. Subjects were also queried about the occurrence of nonrespiratory side effects such as headache, dizziness, etc.

Training

Subjects were familiarized with the sensory experience, mechanical ventilation, and the rating system on 2 to 4 days prior to fMRI scanning. Guided by the airway pressure and flow waveforms, we coached the subjects to relax their respiratory muscles and not fight or assist the ventilator. The adequacy of relaxation during mechanical ventilation was assessed by examining the breath-to-breath consistency of expiratory airflow and inspiratory airway pressure traces as previously described (Colebatch et al. 1991). End-tidal P_CO2 and V_T were varied over a range sufficient to expose the subjects to the full range of air hunger intensities including Extreme. (Stimuli producing Extreme air hunger were avoided in the scanner experiments.) The subjects were encouraged to settle on their own personal use of the rating system for consistency later in the scanner trials.

fMRI scanning and analysis

Imaging was performed using a Siemens Vision magnetic resonance imaging (MRI) scanner, operating at 2 T, with a gradient booster system and local gradient head coil. After an initial positioning image was obtained, a T1-weighted "structural" MRI of each subject's brain was acquired. Subsequent T2* BOLD-sensitive, whole-brain images were collected every 5.2 s; each consisted of 56 sequential axial planes, with 3 mm isotropic voxel resolution and 64 × 64 pixel matrix. The BOLD signal mainly reflects decrease in deoxyhemoglobin concentration due to the large increase in local blood flow that accompanies increased synaptic transmission (Logothetis et al. 2001).

Image manipulations and statistical analyses were performed using Matlab 5 (Mathworks, MA) and SPM99 software (Wellcome Dept. of Cognitive Neurology, Institute of Neurology, London; http://www.fil.ion.ac.uk/spm). To account for spin saturation effects, the first five images were discarded. Image processing began with spatial realignment of each subject's scans to the sixth image of the series. This process ensures spatial congruency and removes minor movement-related artifacts. Each subject's structural data were then spatially normalized to standard stereotactic space based on the Montreal Neurological Institute (MNI) database (Friston et al. 1995a). The normalization parameters were then applied to all functional images for that subject. Images were then spatially smoothed with a three-dimensional 6-mm full-width half-maximum isotropic Gaussian kernel filter to improve S/N ratio.

Statistical tests were performed for the group and for each individual to determine regional BOLD signal changes significantly related to the breathing tasks. For the primary experiment, air hunger ratings reported each breath served as the independent input variable for the SPM fMRI time-series analysis (refer to Fig. 1 for an example of breath-to-breath recording of ratings versus physiological variables). These data were convolved with a hemodynamic response function to represent the relationship between neural activity and changes in cerebral blood flow (Friston et al. 1995b). In the control experiment at low P_CO2, where no air hunger was reported, an ON-OFF function representing V_T was used as the input function. The moving average was convolved with a hemodynamic response function. All data were then temporally smoothed with a high-pass filter (180 s) to remove signal drift and a low-pass filter (6 s, Gaussian) to remove high-frequency noise (Holmes et al. 1997). Ruse V_T transients and global BOLD signal intensity (calculated for the whole brain) were included in the analysis as confounding regressors.

For the statistical analysis, the data from the two experiments were combined to perform a voxel-by-voxel analysis of covariance based on the general linear model (Friston et al. 1995b) as implemented in SPM 99. In this analysis the experiments were represented as a factorial design (Frackowiak et al. 1997a; Friston et al. 1996) with one main effect for each experiment. From the primary experiment, the main effect is the increase in air hunger, combined with the decrease in V_T with a constant, normal PET_CO2; from the control experiment, the main effect is the decrease in V_T in the absence of air hunger with a constant, low PET_CO2. The positive interaction in the analysis, therefore, represents the increases in BOLD signal change associated with air hunger that are independent of any BOLD signal changes that might be produced by the decreased V_T. (It should be noted that the activation map for the main effect of the primary experiment was nearly identical to the map of the interaction analysis, suggesting that the activations seen in the primary experiment were mainly the result of air hunger per se.) Regions of condition-related differences in fMRI signal intensity were represented by maps of the T statistic generated by SPM (threshold P < 0.05 corrected for multiple comparisons), with local maxima reported in x, y, and z reference coordinates (MNI). Reported anatomical loci were determined on the basis of topography of high-resolution (T1-weighted) structural scans using published brain atlases (Duvernoy 1991; Schmahmann et al. 1999; Talairach and Tournoux 1988). Group activations were located using the group average structural scan and individual activations were located using each individual's structural scan.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Psychophysical and physiological

A typical subject's perceptual air hunger response to changes in V_T during the primary experiment is presented in Fig. 1. Our intervention was effective in evoking substantial changes in air hunger: for the group, median air hunger during low V_T epochs was Moderate (50% of the available scale). Air hunger ratings returned to zero during subsequent high V_T epochs (Fig. 2). Consistent with the expected dynamic response, air hunger ratings did not plateau during the 42-s epochs. During the control experiment, none of the subjects reported air hunger during typical low V_T epochs (although 4 of the 6 subjects reported mild air hunger during the very low V_T ruse epochs not included in analysis).

View larger version (15K): [in this window] [in a new window]   Fig. 2. Average time course of air hunger response to tidal volume change. Group (n = 6) results averaged for primary experiment (mean end-tidal PCO2 = 41). Breath by breath mean tidal volumes (VT) and air hunger ratings calculated from 5 epochs in each subject; ruse epochs were excluded from graph. Breaths were aligned with respect to the step change in tidal volume.

When instructed to "describe the respiratory sensations you felt when you rated moderate or severe," subjects typically reported "breathless," "need for more air," "like breath-hold," and "hunger for oxygen." The descriptors chosen from the set list are shown in Fig. 3. From a list of 17 nonrespiratory descriptors, all subjects endorsed a sensation of "restlessness." Each of the following descriptors were chosen by two subjects: "headache," "flushed," and "irritable." The descriptors "heart pounding," numbness, and dizziness were each reported by one subject while none of the subjects reported disorientation, sweating, stomach ache, salivation, coldness, unusual taste/smell, visual, or hearing effects.

View larger version (39K): [in this window] [in a new window]   Fig. 3. List of descriptors. Bars show the frequency with which descriptors were chosen as one of the 2 best terms to describe the sensation. Even when subjects were allowed to choose as many descriptors as they wished, they rarely chose descriptors 2, 3, 9, 11, or 12 (0-3% of choices made).

Brain imaging results

Group results are presented in Figs. 4-6 and Table 1. The fMRI signal was positively correlated (P < 0.05 corrected) with air hunger ratings in several regions often classified as limbic, paralimbic, or phylogenetically old. Bilateral activation occurred in the following regions: anterior insula, pars opercularis, anterior cingulate gyrus, amygdala, putamen, caudate, and cerebellum. Unilateral activations were observed in the right thalamus, left lingual gyrus, and cuneus.

View larger version (98K): [in this window] [in a new window]   Fig. 4. Statistical map showing significant regional signal increases associated with periods of air hunger in main experiment after accounting for effects of tidal volume seen in control experiment (group result, n = 6). T statistic values based on an explicit test of intersession effects (interaction analysis). The glass brain projections are normalized to stereotactic space (see METHODS) in sagittal, coronal, and axial planes. Signal intensity is represented by an arbitrary gray scale (light gray indicates threshold T = 5.1, P < 0.05 corrected for multiple comparisons, darker shades indicate increasing T score). Relevant local maxima are labeled AC, anterior cingulate; In, insula; IPS, intraparietal sulcus; SMA, supplementary motor area; V, cerebellar vermis.

The spatial extent of activation in the anterior insula was notable, comprising distinct superior, mid, and inferior loci. These activations span anatomical locations corresponding to all three cytoarchitectural divisions of anterior insular cortex, from granular (superior) field, through transitional-dysgranular to agranular (inferior) field (cytoarchitectonics reviewed by Augustine 1996; Mesulam and Mufson 1985). Near each of the activations in right anterior insula were activations in frontal regions with confluent cytoarchitecture (i.e., inferior frontal gyrus, vertical ramus, posterolateral orbital frontal cortex; Fig. 5).

View larger version (193K): [in this window] [in a new window]   Fig. 5. Statistical maps of significant brain activation associated with periods of air hunger for the group (same T statistic map as in Fig. 4) superimposed onto the group mean structural image (n = 6). Selected axial slices (z-plane; z = 14 to +14) are displayed illustrating the effect size of regions of interest: insula, prefrontal cortex, basal ganglia, and amygdala. "L, A" indicates left anterior. Horizontal slices are depicted on a coronal image (y = +18) and sagittal image (x = +34). Signal intensity is represented by inset arbitrary color scale (T = 5.1, P < 0.05 corrected for multiple comparison). Relevant local maxima are labeled: Am, amygdala; C, caudate; Fi, inferior frontal gyrus; Fo, orbital frontal gyrus; Ia, insula (agranular); Id, insula (dysgranular); Ig, insula (granular); Op, operculum; P, putamen; T, thalamus; Vr, vertical ramus lateral fissure.

Extensive activation was also observed in the anterior cingulate cortex (spanning >650 voxels; Table 1, Figs. 4 and 6). The anterior cingulate activation was diffuse, yet local maxima were in two distinct clusters, one immediately superior to the corpus callosum genu, and a second extending toward the anterior supplementary motor area (pre-SMA).

View larger version (59K): [in this window] [in a new window]   Fig. 6. Statistical maps of significant brain activation associated with periods of air hunger for the group (same T statistic map as in Fig. 4) superimposed onto the group mean structural image (n = 6). Selected sagittal 4-mm slices (x-plane; x = 4 to +4) are displayed illustrating the effect size of regions of interest: pre-SMA, anterior cingulate, and cerebellar vermis. Vertical slices are depicted on a coronal image (Z = +22). Signal intensity is represented by inset arbitrary color scale (T = 5.1, P < 0.05 corrected). Relevant local maxima are labeled: AC, anterior cingulate; preSMA, presupplementary motor area; V, cerebellar vermis.

Distinct hemispheric cerebellar activations were localized to the quadrangular (VI), inferior semilunar (crII), and biventer (VIIIA) lobes. In addition there was midline activation of the cerebellar vermis, with distinct loci in the declive (VI), tuber (VIIB), and pyramis (VIIIB) lobes (locations according to atlas of Schmahmann et al. 1999).

In addition to the above phylogenetically old regions, fMRI signal increased in neocortical regions associated with motor planning and control. Bilateral activation occurred in the following regions: intraparietal sulcus, premotor cortex, pre-SMA, and sensory cortex. Activation of the prefrontal cortex was multi-focal, with local maxima in the superior frontal (including the premotor area), middle frontal, and inferior frontal (extending to frontomarginal) gyri.

Examination of activations in individual subjects (using a threshold of P < 0.001 uncorrected for multiple comparisons) revealed activation of the anterior insula, anterior cingulate, and prefrontal cortices in all six subjects. Although all subjects demonstrated activation in these structures, a conjunction analysis failed to show statistical significance because of minor variation in anatomic location. Fewer subjects had activations in the intraparietal sulcus and pre-SMA (5 subjects), cerebellum and basal ganglia (4 subjects), and amygdala (3 subjects).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Experimental interventions

STIMULI AND SENSATIONS. In the primary experiment we evoked significant air hunger in the test condition and effectively relieved the air hunger in the baseline condition. The secondary experiment provided an effective control for activations produced by tidal volume changes in the absence of air hunger. A number of studies have shown that mildly elevated CO₂ combined with restricted ventilation evokes a strong sense of air hunger, unaccompanied by other prominent sensations (Banzett et al. 1989, 1990, 1996; Gandevia et al. 1993; Opie et al. 1959). Lowering V_T at constant P_CO2 produces sensations nearly indistinguishable from the sensations produced by raising P_CO2 at constant V_T; these stimuli are often indistinguishable to subjects (Manning et al. 1992). The quality of the sensation the subjects were instructed to rate was based on debriefing results from prior experiments (e.g., Banzett et al. 1996, 2000). As in those studies, our subjects chose the terms "urge to breathe," "like breath-hold," or "starved for air" to describe the respiratory sensations they felt.

MOTOR IMPLICATIONS OF STIMULI. While respiratory discomfort was the most important feature of the present experiment and the three other imaging studies of dyspnea, all of the comparisons have some motor implications as well. In the present study, and our prior study, subjects were required to suppress spontaneous respiratory motor activity to cooperate with the mechanical ventilator. This task requires considerable inhibitory effort during periods of high air hunger. This task may have demanded greater involvement of motor planning areas due to the rapidly changing conditions. In contrast, a higher respiratory motor output accompanied dyspnea in the studies from other laboratories (Brannan et al. 2001; Liotti et al. 2001; Parsons et al. 2001; Peiffer et al. 2001); this can be inferred from the higher airway pressure or higher ventilation. It is difficult to say whether this added activity was generated volitionally or reflexively. Because the motor implications of these various studies were so different, it is unlikely that the activations that are common to all studies have to do with motor acts. However, it is possible that in all these studies subjects were suppressing nonrespiratory motor acts such as removing the mouthpiece, arguably the most natural response to these aversive stimuli.

Activated brain regions

ANTERIOR INSULA. The present results confirm our original observation of prominent activation of the anterior (agranular) insula during dyspnea (Banzett et al. 2000). Our hypothesis that the anterior insula is essential to the perception of dyspnea has been strengthened by observation of insular activation during several imaging studies of dyspnea (Banzett et al. 2000; Brannan et al. 2001; Liotti et al. 2001; Peiffer et al. 2001). Electrophysiological and anatomic tracer investigations have provided evidence linking the insula to afferents and motor centers relevant to breathing. Afferents from respiratory chemoreceptors and pulmonary stretch receptors project to the granular and dysgranular insula, neighboring the principal activation we observed (Hanamori et al. 1998). Stimulation studies of the vagus and the insula have demonstrated reciprocal respiratory projections in man and other mammals (Kaada 1951; Radna and MacLean 1981). In addition, medullary respiratory neurons project to both the granular and the agranular insular cortex (Gaytan and Pasaro 1999); thus our finding is consistent with a leading hypothesis for the generation of air hunger by corollary discharge from brain stem motor activity (reviewed by Banzett and Lansing 1996). The insula has efferent and afferent connections with all of the neighboring limbic and paralimbic structures activated in the present study: operculum, anterior cingulate, orbital frontal cortex, thalamus, amygdala, and basal ganglia (as reviewed by Augustine 1996; Mesulam and Mufson 1982a,b). Further, the insula has more distant connections, projecting to the SMA as well as receiving afferents from the somatosensory, prefrontal, and posterior parietal cortices (all activated in the present study).

AGRANULAR EXTENSIONS (ORBITAL FRONTAL CORTEX, VERTICAL RAMUS, AND FRONTOPARIETAL OPERCULUM). The insular activation associated with air hunger in the present study was not confined to the strong activation in the anterior region: loci appeared elsewhere within the insula "proper," spanning its inferior to superior limits (Fig. 5). In addition, significant loci were identified within paralimbic regions known to be confluent with the cytoarchitectural fields of the insula: vertical ramus, frontoparietal operculum, and orbital frontal cortex. These regions have been studied extensively in rats and monkeys and shown to have reciprocal connections to the insula (Augustine 1996; Cechetto and Saper 1990). Opercular activity was observed at lower significance in two prior studies (Banzett et al. 2000; Peiffer et al. 2001). Because the operculum has contiguous cytoarchitecture and strong connections to the insula, it may be similarly involved in dyspnea. These agranular extensions of insular cortex have been described as having dense connections to nearly all limbic structures, serving emotive and visceral sensorimotor functions (Mesulam and Mufson 1985; Ongar and Price 2000). Further, studies in animals suggest the orbital prefrontal cortex plays a role in guiding emotional behaviors, by shaping primal survival strategies.

CINGULATE GYRUS. We observed strong activation of the anterior cingulate cortex in all subjects, contiguous with the pre-SMA. Most prior studies of dyspnea have not shown cingulate activation. In contrast, nearly all studies of pain do show activation of anterior cingulate, as do studies of thirst and hunger (Casey 1999; Denton et al. 1999; Tataranni et al. 1999). In addition to its proposed role in primal sensation, anterior cingulate cortex is also well known to participate in cortical attentional networks; controversy remains regarding its role in the perception of pain (Derbyshire et al. 1998; Peyron et al. 1999; Rainville et al. 1997). Its appearance in the present experiment may reflect the constantly changing stimulus state.

AMYGDALA. Activation of the amygdala in most of our subjects, and in the group studied in another laboratory (Brannan et al. 2001; Liotti et al. 2001), may relate to the aversive aspects of dyspnea. The amygdala is widely thought to be involved in fear, anxiety, avoidance behavior, and general emotional reactivity (e.g., Adolphs et al. 1995; LeDoux 1992; Morris et al. 1998). Also, functional and anatomic evidence supports a respiratory role of the amygdala in laboratory animals (Davis 1997; Gaytan and Pasaro 1999). Similar to the insula, the amygdala has numerous connections to other limbic and paralimbic regions including insula, anterior cingulate, prefrontal cortex, and thalamus (Davis 1997; Mesulam and Mufson 1985).

CEREBELLUM. The cerebellum was active in most of the present subjects and in both earlier studies of dyspnea that imaged this structure (Parsons et al. 2001; Peiffer et al. 2001). The cerebellum is thus the second most consistently observed activation in studies of dyspnea. Despite the traditional view confining the role of the cerebellum to motor coordination (Houk and Wise 1995), new evidence suggests it has a role in primary sensory processes as well (Damasio et al. 2000; Dolan 1998; Schmahmann and Sherman 1998). Recent studies have demonstrated that the cerebellum, particularly the vermis, is involved in pain (reviewed by Casey 1999) and homeostatic functions such as thirst and hunger for food (Parsons et al. 1999; Tataranni et al. 1999). The vermal and posterior quadrangle activations identified by the present study are considered phylogenetically ancient and may be essential to primal emotion and vigilance functions (Parsons et al. 2001). The role of the cerebellum in respiratory motor control has also been recently shown in both humans and animals (Corfield et al. 1995; Fink et al. 1996; Harper et al. 1998; Xu and Frazier 1997; Zhang et al. 1999). Moreover tracer studies in rats have demonstrated respiratory medullary afferents and efferents projecting to the cerebellum (Gaytan and Pasaro 1999).

FRONTOPARIETAL NETWORK. The present study identified prominent activation of the posterior parietal cortex as well as activations in prefrontal and premotor cortices, primary components of the frontoparietal network. This network is involved in motor planning and attention (Frackowiak et al. 1997b; Mesulam 1990; Passingham 1993; Posner and Dehaene 1994). This network has not been seen consistently during dyspnea, and its activation may be unique to the experimental protocol.

SMA AND PREMOTOR CORTEX. Activation of the pre-SMA and premotor cortex may have been related to the requirement that our subjects consciously suppress the urge to breathe while on the ventilator. Analogous activations were seen in a number of subjects in our earlier study employing the same stimulus paradigm (Banzett et al. 2000). Hsieh et al. found similar activations during itch and proposed they were related to the suppression of the urge to scratch (Hsieh et al. 1994). The coordinates of pre-SMA, premotor area, and middle frontal gyrus observed in the present study are comparable to those activated during "go/no-go" (movement inhibition) protocols (Kawashima et al. 1996). Electrophysiological animal studies, human electroencephalography, and magnetoencephalography suggest the prefrontal cortex is essential to the inhibition of movements (as reviewed by Kawashima et al. 1996; Konishi et al. 1999). SMA and prefrontal activity were absent from the other imaging studies of dyspnea, perhaps because subjects in those experiments did not need to suppress respiratory efforts, but also because the structures were not always in the image field. Notably, the present study and our prior study showed no activation in primary motor cortical regions associated with volitional inspiration (Mckay et al. 2000; Ramsay et al. 1993a).

Limitations of technique

First, techniques that record neural activity provide only one piece of the evidence necessary to determine the function of active neurons. Unequivocal conclusions require parallel evidence from lesion deficits or direct stimulation studies.

Second, current functional imaging techniques (PET and fMRI) do not directly measure neuronal activity, but depend on physiological changes that accompany neuronal activity (predominantly increased regional blood flow). Therefore inferences from BOLD signal may be confounded by physiological changes unrelated to neural activity. One such confound could arise from nonuniformity in the global cerebral blood flow decrease induced by the lower prevailing PCO₂ in the control experiment. We used a factorial analysis that contrasted signal change from baseline in the main (high CO₂) experiment with signal change from baseline in the control (low CO₂) experiment. Thus nonuniform changes in blood flow due to CO₂ could not confound local changes due to neural activity.

Another confound could arise if hypocapnia attenuates the BOLD response to a given sensory stimulus (i.e., if the signal changes due to CO₂ and neural activation do not add linearly). If true, this would have caused the effect of tidal volume to be underestimated in our control experiment. The interaction of neural stimulation with CO₂-induced signal change has been studied in primary visual cortex. One recent study by Posse et al. (2001) found that a given stimulus produced a smaller response at lower PCO₂. This study measured visual response only during the first 20 s of stimulation. During the 30 s following stimulus onset, variable dynamic events occur depending on the stimulus onset characteristics and the details of imaging procedure; for instance, there may or may not be a large overshoot in BOLD signal (Hoge et al. 1999b). It is not known whether the results obtained by Posse et al. apply only to the first 20 s after stimulation. Two other studies (Corfield et al. 2001; Hoge et al. 1999a) suggest that the effect of CO₂ and prolonged visual stimulation add linearly; these studies used longer visual stimuli, more comparable to the slow respiratory stimuli used in the present study. The weight of evidence therefore suggests that this physiological confound is unlikely to explain our principal findings; however, it cannot be completely excluded.

BOLD fMRI is limited by S/N ratio; the technique is susceptible to signal confounds induced by brain motion and motion related to the cardiac and respiratory cycles. BOLD imaging of brain stem structures is thought to be particularly susceptible to movement related artifact (Poncelet et al. 1992); however, distinct medullary respiratory nuclei have been successfully identified by BOLD (Mckay et al. 2000). The failure of the present study to identify brain stem nuclei could be related to limitations of poor S/N; however, inhibition or inactivity of these nuclei during our experimental intervention cannot be excluded.

Summary

The most salient event in our experiment was a strong sense of air hunger. Air hunger, similar to hunger for food, thirst, and the need to escape from pain, is a powerful and primal sensation alerting the organism of a threat to survival. The limbic/paralimbic system, which includes the insula, cingulate gyrus, and amygdala, is thought to aid survival by integrating behavior with the perception of physiological needs (Adolphs et al. 1995; Allen et al. 1991; Ongar and Price 2000). Human neuroimaging studies have shown activation of the cerebellum, limbic, and paralimbic structures in response to essential survival stimuli: pain, hunger, thirst, and dyspnea (Casey 1999; Corfield et al. 1995; Liotti et al. 2001; Parsons et al. 1999, 2001; Peiffer et al. 2001; Tataranni et al. 1999). To what extent are these limbic/paralimbic activations a response to general discomfort? As one example, the anterior insula may serve only as a nonspecific "alarm center" for physiological threat (Reiman 1997), or it may be that there are specific insular neurons activated by each of these stimuli. It will require higher resolution techniques to determine whether air hunger activations in the insula are identical, or different but closely adjacent to, pain or thirst activations. Clearly, the overall pattern of response of the network is different; for instance, air hunger increased activation in the amygdala, while somatic pain causes no change or decreased activation in this structure (reviewed by Peyron et al. 2000).

The present study is the first to investigate the neural correlates of dyspnea with fMRI. The fMRI technique provided enhanced spatial resolution enabling the differentiation of distinct loci in critical regions of interest. The anterior insula was the only activation associated with dyspnea in all three prior PET studies and the present fMRI study. Other activated areas such as operculum, anterior cingulate, orbital frontal cortex, and amygdala are known to have direct neural connections to the insula, suggesting a network functioning with the insula to mediate dyspnea perception (Allen et al. 1991; Andersen 1995; Gaytan and Pasaro 1999). The insula appears to be central in dyspnea perception, but other elements may also be important, and may vary with the quality of dyspnea and with the nature of the accompanying behavioral response. Determination of which activated structures are essential to normal perception awaits other lines of evidence, such as lesion-induced deficits.